Controlling in the Age of AI: More Data Does Not Mean More Control

by Andreas Seufert / Matthias von Daacke

Dashboards, automated forecasts, and modern BI systems are now standard in many companies. But transparency alone does not equate to the ability to manage. This is precisely where one of the greatest challenges for controlling lies in the age of artificial intelligence.

Traditionally, the strength of management accounting has not lain in providing numbers. What has been crucial is the ability to assess their economic significance: Which variance is truly relevant? Which metric indicates a problem—and which is merely statistical noise? And which management decision would be premature based on this information?

The BI Gap: When Accurate Numbers Lead in the Wrong Direction

Modern BI systems can consistently define, visualize, and automatically deliver data. Nevertheless, control errors often arise not from incorrect data, but from misinterpretation.

Here’s an example: A company’s overall margin is declining. At first glance, profitability appears to be worsening. However, if you look at the individual segments, a different picture may emerge: Product margins are rising, but at the same time, the revenue mix is shifting toward a segment with lower margins.

The figure is correct. The conclusion may still be wrong.

This highlights a fundamental problem: The average describes a result—but the reality of management often falls short of that.

Four Typical Misjudgments

Four common errors in reasoning often arise on the path from data to decisions:

1. Accurate data does not necessarily equate to a common language for decision-making.
   Sales, controlling, and management may look at the same metric and still assign different economic meanings to it. A “single source of truth” therefore does not automatically solve the problem of shared understanding.
2. Aggregations do not always reflect operational reality.
   Overall figures can mask opposing trends in individual segments. Those who focus exclusively on aggregated KPIs risk managing symptoms rather than causes.
3. A driver tree does not automatically explain a cause.
   Breaking down a key performance indicator is mathematically correct, but it does not necessarily answer the question of the economic driver. Declining volumes, for example, can result from weaker demand—or from a lack of delivery capacity, product availability, or a shift in the customer mix.
4. A forecast is not automatically a reliable picture of the future.
   A model is only as good as its data, assumptions, and logic. If flawed definitions or untested cause-and-effect relationships are incorporated, the forecast merely carries existing errors in reasoning forward into the future.

Why AI Doesn’t Automatically Solve These Problems

The hope that artificial intelligence will eliminate the weaknesses of traditional BI falls short. AI can recognize patterns, generate explanations, and prepare decisions. However, it does not automatically correct unclear control logic.

On the contrary: The more analysis is automated, the more important it becomes to be able to critically evaluate its results.

This also changes the role of management accounting. In the future, the key skill will lie less in preparing reports and more in questioning assumptions, cause-and-effect relationships, and decision-making logic.

Business partnering begins with sound judgment

In the age of AI, business partnering therefore does not mean implementing as many new technologies as possible. It means enabling better decisions.

The crucial question is not:

“What AI should we implement next?”

Rather:

“Which decisions are better justified by this AI?”

Data literacy forms the foundation for this. Controllers must not only be able to read numbers, but also understand their meaning, limitations, and potential biases. Only then can they build on this foundation with additional skills such as pattern recognition, root cause analysis, and, ultimately, the use of analytical AI.

This is because AI can speed up the analysis. However, humans remain responsible for the quality of the control logic.